Applied AI Pod
NLP, Speech Tech, Transformer Models, w/ Marc von Wyl, Algolia, E30
Episode Summary
Join a conversation with Marc von Wyl, Sr ML Engineer @Algolia. Marc is teaching Natural Language Processing @EPITA, and is an experienced computer scientist specialized in Natural Language Processing, Machine Learning, and languages in general. Together, we dig into: unstructured data, improving error and ambiguity, and future of NLP.
Episode Notes
- 01:15 - How does NLP work?
- 04:05 - How do Transformer-based NLP models work?
- 08:20 - How to look at unstructured data to take advantage of it more.
- 12:00 - How to leverage ML to bring more to unstructured data?
- 15:25 - Approach for low resources languages.
- 23:25 - Word embeddings for common reasoning needs.
- 26:55 - Techniques to follow to improve error and ambiguity in training data or for a model in general.
- 30:10 - Are GPTs leading effort in the field in a wrong direction?
- 34:15 - Is DeepLearning the end of AI?
- 37:20 - What are some good NLP metrics to watch?
- 42:05 - How do we get past transactional queries to conversational queries?
- 52:00 - Is the Turing test still relevant for NLP or has it become obsolete?
References: